Biological neural networks operate with significantly lower energy costs than their artificial counterparts. Harnessing this efficiency could drastically reduce the massive energy consumption of artificial intelligence systems. One method of doing this involves using biological neural networks grown on micro-electrode arrays (MEA) in vitro as reservoirs within reservoir computing systems. However, involving biology in computing does not come without challenges. In vitro neural networks develop and change over time, which may affect their representation of, and dynamic range to, input. This can be problematic when an input encoder is calibrated to a given dynamic range, and a readout decoder is trained based on a set of representations that are no longer produced by the network. In this study we explore the stability of input representations and dynamic range in biological neural networks. We use a series of input consisting of inter-stimulation intervals to determine the networks’ dynamic range to stimulation timing at different days in vitro. To assess the stability of the representations of input we also train classifiers based on the response on one Day In Vitro (DIV) and test on responses from two other days in vitro. Our results show that networks are generally unstable both in dynamic range and in the stability of their representations.